Image Image Image Image Image Image Image Image Image Image

Bartlett School of Architecture, UCL

Scroll to top


No Comments

Gears and Alexa

  • On February 21, 2018

Still following the earlier approach of working simultaneously in different directions, I feel a bit overwhelmed. But on the other hand, I am learning a lot. Let’s talk about each of them…

Reinforcement Learning

Cartpole is one of the most traditional RL problems. The idea is to teach an agent to balance a pole vertically over a bar. I used ‘random-search’ and ‘hill-climbing’ algorithms to build this. They are very well described here. Coded from the scratch, I feel more and more confident with every new line of code. Check out the video (on top) and the code on

Last time, I used Q-learning to build a maze solving agent. Next is to build a neural network with at least one hidden layer. And probably start moving away from Processing… The bigger idea behind this is to build robots (could be any scale) which can survive, and thrive in environments around human. So every time they enter a new environment, or meet new users, they must adapt themselves accordingly. More on that later.

Alexa-Arduino competition

I started working on this competition 2 weeks back and this is definitely one of the most complicated things I have tried. The idea is to use Amazon Alexa and Arduino to build a fully working IoT gadget for home.

My proposal is a simple shelf, which has multiple containers. It reveals the container which is relevant for users at that point in time. For example, it will show books while you are studying, but beer bottles while you are chilling out. Ideally, it will use reinforcement learning to do so. Predict human behaviour and respond accordingly.

However, for the sake of this competition, I am trying to hook it up with an Alexa. And it is not easy. It is a complex and messy world. Every time I try to go through the process, I end up being frustrated, tired and lost. Here are a few links if you are interested…

However, I feel that I am almost through and today hopefully, is the lucky day. If I am able to do this, it’ll essentially be a fully working IoT device, ready to be shipped… Yes, it is a big deal !!!

3 and half days until the deadline. Wish me luck !!!

Fresh laser cuts, such cuteness…

How not to design ball bearings

Edge of Chaos

Edge of Chaos is a moving installation designed at Interactive Architecture Lab. Ruairi and Vasilija are leading the project, and it’s super ambitious.


I have been lucky to get a chance to work on the installation. My task is to figure out the electronics… PCB design, components, bill of materials, etc… I have always been scared of the ‘hardcore’ electronics. But for what I want to do, this is a crucial skill. Sean and Ruairi have been guiding me through this process, and I recently finished my first PCB design, and it has already been sent to manufacturing. Hopefully, it’ll work !!! Fingers crossed.

Edge of Chaos board designed v1. Proudly designed by me.

Submit a Comment